Research Data and Code for Knowledge Indexing and Structured Analysis Methods for Mitigating Hallucinations in Large Language Models

It includes:Source Code & Algorithms: Implementation of the BERT-CRF domain concept indexing, head-concept-driven dynamic window relation extraction, and TextRank centrality calculation scripts.Experimental Prompts & Configs: Structured Prompt templates and Structured Retrieval-Augmented Generation (RAG) configuration files used in Section 4.2 for GPT-4o (temperature=0, max_tokens=512).Dataset Samples: Expert-annotated concept indexing and relation triplet samples in the diabetes domain (including schema descriptions for 1,458 training and 369 validation samples).